Rational Approach to Identify RNA Targets of Natural Products Enables Identification of Nocathiacin as an Inhibitor of an Oncogenic RNA

Abstract

The discovery of biofunctional natural products (NPs) has relied on the phenotypic screening of extracts and subsequent laborious work to dereplicate active NPs and define cellular targets. Herein, NPs present as crude extracts, partially purified fractions, and pure compounds were screened directly against molecular target libraries of RNA structural motifs in a library-versus-library fashion. We identified 21 hits with affinity for RNA, including one pure NP, nocathiacin I (NOC-I). The resultant data set of NOC-I-RNA fold interactions was mapped to the human transcriptome to define potential bioactive interactions. Interestingly, one of NOC-I’s most preferred RNA folds is present in the nuclease processing site in the oncogenic, noncoding microRNA-18a, which NOC-I binds with low micromolar affinity. This affinity for the RNA translates into the selective inhibition of its nuclease processing in vitro and in prostate cancer cells, in which NOC-I also triggers apoptosis. In principle, adaptation of this combination of experimental and predictive approaches to dereplicate NPs from the other hits (extracts and partially purified fractions) could fundamentally transform the current paradigm and accelerate the discovery of NPs that bind RNA and their simultaneous correlation to biological targets.

Introduction

Natural products (NPs) have acquired functional activities over their producers’ millions of years of Darwinian evolution.(1) Selective pressure on the constitutive organisms that produce NPs has resulted in a myriad of chemotypes encoded by their genomes. Human health has been significantly impacted by NPs, as they have informed the design of ~65% of small molecules approved as medicines by regulatory bodies such as the Food & Drug Administration (FDA) from 1981 to 2019.(2−4)

Biomedical applications of NPs traditionally rely on function-centric approaches, most commonly consisting of phenotypic screening, bioactivity-guided isolation, and subsequent studies to determine the molecular target and mechanism of action (MOA).(5,6) This process is often carried out in a linear fashion, resulting in extensive time and effort to first dereplicate and identify the NP and then to correlate the molecule with its biological target. To overcome some of these challenges, direct screening of NPs on defined, biologically relevant targets has been pursued.(4,7,8)

Various methods have been developed to screen for small molecules that bind RNA.(9,10) Our laboratory has developed a selection-based platform, named Inforna, that enables the design of RNA structure-specific small molecules from sequence, affording chemical probes of RNA function.(11) Given the proven track record of NPs in the clinic and the success of Inforna in targeting RNA, combining the two could enrich our understanding of molecular recognition as well as diversify the known chemotypes with affinity for RNA. The Natural Products Discovery Center (NPDC) at Scripps Research houses over 125 000 bacterial strains, including both our in-house collection and the historic pharma collection that Pfizer contributed to Scripps Research in 2018.(12) The potential number of encoded NPs, exquisitely decorated by nature to elicit potent and unique biological activities, awaiting discovery from this strain collection is immense. The NPDC also contains a collection of 46 031 crude extracts and 28 739 partially purified fractions, made from 14 635 actinobacterial strains,(12) as well as 650 pure NPs. Herein, a rational approach to identify RNA targets of NPs is described, highlighting the identification of nocathiacin I (NOC-I) as an inhibitor of the oncogenic, noncoding microRNA (miR)-18a.

Results and Discussion

Discovery of NPs That Bind RNAs by Two-Dimensional Combinatorial Screening

To identify NPs and their RNA-binding partners, a library-versus-library selection, dubbed two-dimensional combinatorial screening (2DCS),(11) was undertaken. This approach studies the binding of thousands of RNA folds present in the human transcriptome to NPs that are noncovalently absorbed onto microarray surfaces (AbsorbArray (13)). AbsorbArray was previously validated by studying RNA structures that bind to over 1000 diverse small molecules including those housed in the NIH Clinical Collection, SYNlibrary95 (kinase inhibitors), and an RNA-focused library.(13) These studies showed that in the case of mitoxantrone, nearly all of the small molecule delivered to the surface is maintained, although this is likely small-molecule dependent.(13)

To demonstrate that NPs can bind to RNAs, we constructed a pilot library comprising 11 273 crude extracts and 5082 partially purified fractions, made from our in-house collection (Table S2), as well as 72 (of 650) pure NPs (Figure S1). Partially purified fractions were generated by subjecting a crude extract to reverse phase flash chromatography. This subset of pure NPs was selected based on their physicochemical properties, i.e., they are predicted to be biased for binding RNA by nature of their shared properties with known RNA binders housed in Inforna (Figure S2). For example, compared with FDA-approved drugs (in DrugBank (14)), RNA-targeted small molecules have more hydrogen bond donors (5.24 ± 4.63 vs 2.40 ± 5.00) and acceptors (8.63 ± 4.60 vs 5.12 ± 7.38) and slightly larger topological polar surface area (TPSA; 156.12 ± 118.62 vs 96.10 ± 148.01), not unexpected from chemical differences between proteins and RNA. The physicochemical properties of the selected pure NPs are therefore similar to known RNA binders (Figure S2).

The pilot library was screened for binding to two different RNA libraries. One RNA library displays the randomized region in a 3 × 3 nucleotide internal loop pattern, while the other displays the randomized region in a 3 × 2 nucleotide pattern (i.e., internal loop libraries; ILLs) (Figure 1). The 3 × 3 ILL comprises 4096 (46) unique RNAs that form symmetric internal loops (3 × 3, 2 × 2, and 1 × 1), bulges, and base pairs, while the 3 × 2 ILL comprises 1024 (45) unique RNAs that form asymmetric internal loops (3 × 2, 2 × 1), bulges, and base pairs. These libraries were chosen as they cover a large percentage of structural motifs found in microRNA (miRNA) precursors. We previously cataloged all secondary structures formed by human miRNA precursors, affording 7436 structural motifs that include bulges, internal loops, hairpin loops, and multibranch loops. (15) Of these, 78% are internal loops or bulges, where small motifs such as those found in our two RNA libraries predominate. Specifically, our two RNA libraries capture ~75% of the internal loops and ~70% of bulges found in miRNA precursors. The cassette into which the randomized region was inserted was previously developed to study thermally stable noncanonically paired regions.(16) The ultrastable GAAA hairpin loop was chosen to ensure proper folding of the library. Single-stranded regions on 5′ and 3′ ends allow amplification by reverse transcription polymerase chain reaction (RT-PCR), while the stem was also carefully chosen to enable binding of RT-PCR primers and hence efficient amplification.

Figure 1.

Two-dimensional combinatorial screening (2DCS) selection of the RNAs that bind natural products (NPs). Top, the strain collection and the pilot library, including crude extracts, partially purified fractions, and pure NPs, which were screened for binding 3 × 3 nucleotide and 3 × 2 nucleotide internal loop libraries (ILLs), which led to the discovery of nocathiacin I (NOC-I) as an RNA binder. Bottom, hit rates for different NP sample types from the pilot library of 11 273 crude extracts, 5082 fractions, and 72 pure NPs and mining of the selected RNA motif–small molecule interactions resulting in the discovery of NOC-1 and its lead RNA target (pre-miR-18a).

The screen resulted in the identification of 16 crude extracts, four partially purified fractions, and one pure NP, NOC-I, as hits that bound RNA structural motifs. A sample was considered a hit if the signal was >3-fold than that of the background (Figure S3A); a secondary validation was completed by AbsorbArray affording 20 RNA-binding hits (Figure S3B). Six NP samples (five crude extracts and one fraction) were unique for 3 × 3 ILL, while six other NP samples (five crude extracts and one fraction) only bound 3 × 2 ILL (Figures 1 and S3). NOC-I was the only hit from pure NPs, which showed binding to both ILLs (Figures 1 and S3–S5). (LC-MS analysis showed that NOC-I was not present in the crude extracts or fractions.) The hit rates of crude extracts (0.14%) and fractions (0.079%) from our studies are comparable to the hit rates from other combinatorial (0.7%) or synthetic libraries (0.03%), while that of the pure NPs is higher (1.39%), likely due to the biased RNA-binding properties of the pure NPs tested.(17,18) As NOC-I was the only pure NP, it was selected to showcase that our approach, the hits identified thereof, can lead to the discovery of NPs that bind RNA targets and abrogate cellular functions. As dereplication of crude extracts and fractions is a laborious process, those studies will be the subject of subsequent studies. Notably, the antibacterial properties of NOC-I were discovered from a phenotypic screen of an antibiotic-resistant strain of Enterococcus faecium.(19,20) Its mode of action was subsequently traced to the binding of ribosomal protein L11, as determined by the mutational mapping of resistant strains.(21)

NOC-I as a Proof-of-Concept to Define RNA-Binding Landscape and Mine RNA Targets

To define its RNA-binding landscape, NOC-I was absorbed onto a microarray surface (13) and incubated with radioactively labeled 3 × 3 or 3 × 2 ILL in the presence of a large excess of unlabeled RNA competitors that mimic regions common to all library members and constrain binding to the randomized region as well as 1000-fold excess of bulk tRNAs, as compared to the RNA library (Figure S5). The RNAs selected by NOC-I were harvested from the array and subjected to RNA-seq analysis for identification. By also subjecting the starting library to RNA-seq analysis, RNA structures that are preferred by NOC-I can be defined by calculating their statistically significant enrichment in the selected RNAs.(22) Here, the relative frequency of a motif of interest in the selection is compared to its relative frequency in the starting library by a pooled population comparison that affords the statistical parameter Z_obs, which can easily be converted into a p-value if desired. (22) We have dubbed this method of statistical analysis of selection data High-Throughput Structure–Activity Relationships Through Sequencing or HiT-StARTS. A Z_obs > 1.9 indicates statistical significance equaling a p-value < 0.05. Our previous studies showed that Z_obs correlates with affinity and thus defines the RNA-binding landscape of the compound of interest. (22)

Indeed, NOC-I’s binding landscapes for the 3 × 3 and 3 × 2 ILLs were defined (Figure S6), and its binding affinity for an RNA structure correlated with the statistical significance of its enrichment, i.e., Z_obs (Figure S7). In particular, a selected RNA must have an enrichment with a Z_obs > 2.5 (corresponds to p < 0.006) to bind to NOC-I measurably, with a range of affinities of 11–50 μM for 3 × 3 ILL and 8–22 μM for 3 × 2 ILL. Notably, a Z_obs > 4 is required to bind with an affinity of <50 μM, corresponding to eight RNA 3D folds from 3 × 3 ILL and 11 from 3 × 2 ILL (Figure S7). Analysis of the binding landscape for NOC-I revealed that 10 (6%) are new motifs that previously had no known small-molecule binder (Figure S6D).

These interactions were deposited into Inforna to provide a data set that informs lead small molecules for RNA targets. Notably, our platform technology integrates advances in RNA structure determination and prediction, RNA–small molecule interaction data (all interactions reported in the literature, not only those discovered by 2DCS selection), and the statistical significance of these interactions.(11) The structural elements found in cellular RNA targets are mined against the database of RNA motif–small molecule interactions, where overlap indicates lead molecules and lead targets, including putative binding sites.

We therefore mined the new NOC-I interactions against all human miRNAs or the miRnome. Akin to other types of RNAs, miRNAs are transcribed as primary transcripts (pri-miRNAs), followed by step-wise enzymatic processing by Drosha and Dicer, to form precursor (pre-miRNAs) and mature miRNAs, respectively.(23) The mining results were refined by the following criteria: (i) the miRNA must be disease-associated; (ii) the putative binding site must be in a nuclease processing site. We have previously shown that a small molecule must bind these functional sites to inhibit miRNA biogenesis;(24) and (iii) the statistical significance of the enrichment in the 2DCS selection, i.e., Z_obs. This analysis revealed a promising match between NOC-I and the Dicer processing site of pre-miR-18a, (5′GAU/3′C_A) derived from the 3 × 2 ILL selection (Figures 1 and S6). Interestingly, miR-18a is part of the well-known oncogenic miR-17/92 cluster. (25) In prostate cancer, miR-18a is upregulated and represses the pro-apoptotic protein serine threonine kinase 4 (STK4) (Figure 2A). (26) We therefore investigated whether NOC-I can inhibit miR-18a biogenesis, de-repress STK4, and promote apoptosis in prostate cancer cells.

Figure 2.

NOC-I selectively targets pre-miR-18a and inhibits its biogenesis in the prostate cancer cell line DU-145. (A) In prostate cancer, miR-18a, part of the miR-17/92 cluster, is upregulated and represses pro-apoptotic protein serine threonine kinase 4 (STK4), leading to evasion of apoptosis. Inhibition of miR-18a de-represses STK4, triggering apoptosis in DU-145 cells. (B) Effect of NOC-I treatment on mature miR-18a levels in DU-145 cells, as determined by RT-qPCR (n = 6). (C) Effect of NOC-I treatment on pre-miR-18a levels in DU-145 cells, as determined by RT-qPCR (n = 3). (D) Effect of NOC-I treatment on pri-miR-17/92 levels in DU-145 cells, as determined by RT-qPCR (n = 3). All p-values were calculated using a two-tailed Student’s t-test, where *, p < 0.05; **, p < 0.01; and ***, p < 0.001. Data are reported as the mean ± standard error of the mean (S.E.M.)

NOC-I Binds Dicer Processing Site of pre-miR-18a In Vitro

We first studied the binding affinity of NOC-I for a model of miR-18a’s Dicer processing site containing 5′GAU/3′C_A using a direct binding assay that monitors the inherent fluorescence of NOC-I (Figure S7) as a function of RNA concentration. Indeed, NOC-I binds miR-18’s Dicer site with a K_d of 8 ± 2 μM (Figure S7). NOC-I also binds to pre-miR-18a, with a K_d of 4 ± 1 μM, while no saturable binding was observed to a pre-miR-18a mutant in which the A bulge was converted to an AU base pair, thereby abolishing the putative binding site (Figure S8). In agreement with these binding studies, NOC-I inhibited the in vitro Dicer processing of wild-type pre-miR-18a, with an IC₅₀ of 13 ± 1 μM, but not the mutant (Figure S9). Notably, we previously identified a substituted benzimidazole with steric bulk that inhibited Dicer processing of pre-miR-18a. NOC-I is ~4-fold higher affinity than the benzimidazole inhibitor (Table S3), and a comparison of their binding landscapes indicates that NOC-I should be more selective: NOC-I is predicted to bind 81 RNA motifs in 3 × 3 ILL (Z_obs > 2.5), while the benzimidazole inhibitor is predicted to bind 215 RNA motifs in 3 × 3 ILL (Z_obs > 7, i.e., Z_obs that predicts binding for this molecule) (Figure S10). (22)

NOC-I Selectively Inhibits Cellular Processing of pre-miR-18a in DU-145, Prostate Cancer Cells

As aforementioned, miR-18a is aberrantly expressed in prostate cancer cells, conferring evasion of apoptosis. (26) We therefore assessed the cellular activity of NOC-I in the prostate cancer cell line DU-145, previously validated for evading apoptosis via the miR-18a-STK4 circuit.(26) We first measured the effect of NOC-I treatment on miR-18a biogenesis by measuring pri-, pre-, and mature miR-18a levels by RT-qPCR. If NOC-I’s mode of action is indeed inhibition of miR-18a biogenesis, then a reduction of mature miRNA levels and an increase in pre-miR-18a levels are expected. Indeed, NOC-I reduced mature miR-18a levels in a dose-dependent manner, with reductions of ~40 and ~60% observed at 0.2 and 2 μM, respectively (Figure 2B); pre-miR-18a abundance was increased by ~2.5-fold at the 2 μM dose (Figure 2C), while the levels of pri-miR-18a remained unchanged (Figure 2D). NOC-I had no significant effect on mature miRNA levels of other miR-17/92 cluster members (miR-17, -20a, -19a, -19b-1, and -92a-1) (Figure S11). Importantly, NOC-I activity was ablated upon forced expression of pre-miR-18a or the pri-miR-17/92 cluster (Figure S12).

Many factors could contribute to the observation that the concentration required to inhibit biogenesis (EC₅₀ ~ 200 nM) of pre-miR-18a is ~20-fold lower than its K_d, including active cellular uptake of the compound, cellular localization, particularly colocalization with the target, increased cellular permeability, or a combination thereof. We attempted to study the NOC-I’s localization in DU-145 cells using confocal fluorescence microscopy; however, the compound’s fluorescence was very weak, disallowing us to definitively determine cellular localization.

NOC-I Binds to the Dicer Site of pre-miR-18a in Cells as Measured by ASO-Bind-Map

To validate direct engagement of pre-miR-18a by NOC-I, we employed the target validation method named ASO-Bind-Map.(11) In this method, an antisense oligonucleotide (ASO) competes with the small molecule for binding to the putative binding site within the RNA target (Figure 3A). If the small molecule binds the putative site, then the ASO is unable to hybridize with the target and trigger RNase H cleavage. If the ASO’s and small molecule’s binding sites do not overlap, then the ASO induces targeted cleavage (Figure 3A). In this study, four ASO gapmers (2-O-methoxyethyl [MOE] phosphorothioates) were used: (i) a gapmer complementary to nucleotides 1–14, which includes nucleotides present in a 1 × 1 UU internal loop and a 1 × 1 UC internal loop (“Gapmer 1–14”); (ii) a gapmer complementary to nucleotides 15–28, which overlaps with NOC-I’s putative binding site, the A bulge at the Dicer processing site (“Gapmer 15–28”); (iii) a gapmer complementary to nucleotides 26–39, which is complementary to the hairpin loop and adjacent U bulge (“Gapmer 26–39”); and (iv) a scrambled gapmer control oligonucleotide that shares no sequence complementarity to pre-miR-18a (Figure 3A). When applied individually to DU-145 cells in the absence of NOC-I, all three complementary oligonucleotides induced RNase H cleavage of pre-miR-18a, reducing the RNA’s levels by ~40%; the scrambled gapmer control had no effect (Figure 3B). When DU-145 cells were co-treated with 0.2 μM of NOC-I and Gapmer 1–14 or Gapmer 26–39, pre-miR-18a levels were reduced to a similar extent as observed upon treatment of gapmer alone, indicating that the small molecule and gapmer binding site did not overlap. In contrast, pre-miR-18a levels were partially restored when DU-145 cells were co-treated with Gapmer 15–28 and NOC-I (Figure 3B). These findings support that NOC-I binds pre-miR-18a’s Dicer processing site in cells.

Figure 3.

Target validation of NOC-I in DU-145 cells using ASO-Bind-Map. (A) Schematic of ASO-Bind-Map to study small-molecule binding to RNA targets. In brief, ASO-Bind-Map is a cellular competition experiment between an antisense oligonucleotide (ASO) and a small molecule. If the two compete for the same binding site, then the small molecule inhibits ASO binding and hence RNase H-mediated degradation. The binding sites for ASOs are indicated as well as the miR-18a’s Dicer site (blue circle), NOC-I’s putative binding site. (B) Relative levels of pre-miR-18a upon treatment of the indicated gapmer oligonucleotide in the absence and presence of NOC-I, as determined by RT-qPCR (n = 3). All p-values were calculated using a two-tailed Student’s t-test, where ***, p < 0.001. Data are reported as the mean ± S.E.M.

Downstream Effects of NOC-I in DU-145, Prostate Cancer Cells

As NOC-I inhibited the biogenesis of miR-18a, we next studied its downstream effects, particularly its rescue of the translational repression of STK4 protein and evasion of apoptosis. Indeed, treatment of DU-145 cells with NOC-I increased STK4 protein levels by ~1.4-fold at the 2 μM dose, as determined by Western blotting (Figure 4A). To study if the translational de-repression of STK4 via NOC-I’s inhibition of miR-18a biogenesis was sufficient to trigger apoptosis, we measured Caspase 3/7 activity. Indeed, Caspase 3/7 activity increased in dose-dependent fashion, with a significant boost observed at 0.2 (p < 0.05) and 2 μM (p < 0.001) (Figure 4B). Furthermore, this triggering of apoptosis by NOC-I was ablated upon forced expression of pre-miR-18a or the pri-miR-17/92 cluster, suggesting that the effect on phenotype is primarily driven by the levels of miR-18a (Figure 4B).

Figure 4.

NOC-I affects downstream targets of miR-18a and induces apoptosis in DU-145 cells. (A) Representative Western blot of the effect of NOC-I on STK4 protein levels and quantification thereof (n = 4). (B) Caspase 3/7 activity on NOC-I treatment in DU-145 and in DU-145 cells forced to express pre-miR-18a or the pri-miR-17/92 cluster. (C) Profiling miRnome-wide selectivity of NOC-I, as determined by RT-qPCR (n = 3), represented by a volcano plot. Note that miRNA profiling was completed after only a 12 h treatment with NOC-I to avoid changes observed due to induction of apoptosis (in contrast to the 48 h treatment period used in other experiments reported herein). Only miR-18a levels are significantly downregulated. (D) Global proteomics analysis of NOC-I-treated DU-145 cells (2 μM) (n = 3). (E) Cumulative distribution of miR-18a-5p’s direct protein targets compared to other cluster members (miR-17, miR-20, miR-19a, miR-19b, and miR-92a-1) and to miR-155, an oncogenic miRNA that contains miR-18a’s A bulge harbored in a nonfunctional site. It is also expressed at similar levels as miR-18a in DU-145 cells (0.54-fold). Dotted lines represent a false discovery rate (FDR) of 1% and a group variance of S0 (0.1). All p-values were calculated using a two-tailed Student’s t-test, where *, p < 0.05 and ***, p < 0.001. Data are reported as the mean ± S.E.M.

Selectivity Studies of NOC-I miRnome- and Proteome-wide

We next assessed the selectivity of NOC-I both miRnome- and proteome-wide. In particular, we studied the effect of NOC-I on 374 miRNAs that are expressed in DU-145 cells and found that only levels of mature miR-18a were significantly decreased, by ~40% (p < 0.01; Figure 4C). Interestingly, 14 other miRNAs contain the 5′GAU/3′C_A bulge bound by NOC-I (Table S4). Of these, the A bulge is found in or adjacent to the Dicer or Drosha processing site of five (miR-101–1, 196a, -3945, -4435, and -4454). Despite this, the levels of the corresponding mature miRNAs are not affected as they are expressed at much lower levels than miR-18a (0.02–0.64-fold), except for miR-4454 (2.1-fold), which is not disease associated (Table S4). As expected, the levels of the nine RNAs in which the A bulge is present in a non-functional site are unaffected by NOC-I treatment, irrespective of the expression level (0.02–1.6-fold relative to miR-18a; Table S4). Our previous studies have shown that both the location of the small-molecule binding site (functional vs non-functional) and expression level influence whether a small molecule inhibits a miRNA’s biogenesis. (24) The binding landscape of NOC-I indicates that it also binds to the Dicer processing site of miR-18b (K_d = 18 ± 2 μM; 5′UUA/3′A_U) and the Drosha site of miR-151 (K_d = 11 ± 4 μM; 5′GAU/3′C_A) (Figures S6 and S7). When NOC-I was treated at 2 μM, miR-151 levels significantly decreased by ~50% while miR-18b levels were decreased by ~20%, although this reduction was not statistically significant (Figure S13).

Complementarily, the effect of NOC-I on DU-145’s proteome was also assessed. Of the 3,379 detectable proteins, 21 were significantly upregulated and five were downregulated (LYN, CCDC12, COX7A2, ARMC6, and MRPS26, none of which are targets of the miR-17/92 cluster) (Table S5). Of the 21 upregulated proteins, four are the direct targets of other miRNAs in the miR-17/92 cluster (CDC73, NRBP1, HS1BP3, and DAG1), as predicted by TargetScan Human v7.2, (27) and seven are the predicted direct targets of miR-18a (DAG1, ERGIC2, UBXN7, HS1BP3, SMCE1, CDC73, ATM, the latter of which was also experimentally validated (23)) (Figure 4D). [Some of the proteins upregulated upon NOC-I treatment are also the direct targets of other members of the miR-17/92 cluster. It is not uncommon for mRNAs to be regulated by multiple miRNAs, in particular those in the same cluster or family.(28)] Notably, upregulation of STK4 was also observed in the proteomics data, although it was not statistically significant (Figure 4D). To determine whether changes in the proteome were correlated with the inhibition of miR-18a biogenesis, we calculated fold changes for the protein targets of miR-18a-5p as a function of cumulative distribution, which assesses if statistically significant changes were observed for miR-18a-5p’s downstream targets, as predicted by TargetScan.(27) Indeed, the analysis revealed that miR-18a-5p’s protein targets were upregulated with p < 0.00001 (Figure 4E), indicating abrogation of the miR-18a circuit that operates in the evasion of apoptosis observed due to its overexpression in prostate cancer cells. Of note, no such correlation was observed for the downstream targets of the other miRNAs in the 17/92 cluster (analyzed collectively) or miR-155, an oncogenic miRNA that contains the same A bulge in a non-functional site and is highly expressed in DU-145 cells (Figure 4E).

Canonical Activity of NOC-I, Binding of Bacterial Ribosomal Proteins

As mentioned above, NOC-I’s mode of action is binding of ribosomal protein L11.(21) Intriguingly, this mode of action is different from the related antibiotic thiostrepton, which binds 23S rRNA at the L11 binding site; binding induces ribosome stalling by inhibiting a required 23S-L11 conformational change.(29) Thiostrepton binds the bacterial RNA–protein complex with K_ds in the low nM range,(30) which are significantly lower than the K_d in our study. Although Escherichia coli and human L11 are ~40% identical (31) and human L11 has key proline residues required for the binding of thiopeptides (Figure S14),(29) we did not observe global translational repression in DU-145 cells upon treatment of NOC-1 (Figure 4D), indicating that the human ribosome is not a significant off-target. Rather, we observed the upregulation of proteins regulated by miR-18a (Figure 4E).

Conclusions

In summary, a rational and efficient approach to identify NPs with affinity for RNA was described. In particular, the binding landscapes of NPs were defined using a library-versus-library selection strategy followed by parallel processing to map NPs to their RNA target motifs. NPs in drug discovery are often considered “privileged” scaffolds for different enzymes and receptors, while their effects on RNA biology were rarely reported. Here, we describe an efficient approach to discover RNA-targeting NPs without the time-consuming dereplication and screening process against individual RNA targets. NOC-I, identified in the screen of the pilot library consisting of crude extracts, partially purified fractions, and pure NPs, was selected to showcase that our approach can discover NPs that bind RNAs and abrogate their cellular functions. Indeed, NOC-I binds to the RNA motif 5′GAU/3′C_A, present in pre-miR-18a’s Dicer processing site, and inhibits its biogenesis in a prostate cancer cell line, disrupting the miR-18a-STK4 circuit that allows prostate cancer cells to evade apoptosis. Importantly, miRnome- and proteome-wide profiling showed that NOC-I is selective for this circuit.

RNA-targeting NPs are known; for example, surfactins have been recently reported as inhibitors of pre-miR-21, (32) and aminoglycosides inhibit bacterial translation by binding to prokaryotic ribosomal RNA (rRNA). (33,34) This study highlights 2DCS as a rational approach to identify RNA-targeting NPs, known or new. Indeed, we previously discovered that anti-cancer drugs, presumed to bind protein targets, also bind RNA. In particular, kinase inhibitors bind RNA and inhibit a cancer cell phenotype. (13) Therefore, this study is a step in the same direction to explore the RNA binding of known NPs, which in turn can accelerate the identification of NPs targeting many disease-relevant RNAs.

Collectively, our findings demonstrate that NPs indeed have RNA-binding capacity and affect biology. As NPs possess a wealth of chemical space that is largely unexplored for RNA, the rational approach described herein sets the stage to exploit the unprecedented NP diversity, as exemplified by the 20 additional hits (16 crude extracts and 4 partially purified fractions) already identified from the pilot screen in the current study (Figures 1 and S2). The actinobacterial strain collection and the NP library at Scripps Research therefore may provide a source of rich and unique NP chemotypes to target RNAs. Indeed, various laboratories have shown the importance of RNA targeting, and the identification of NPs that bind RNA will surely increase the structural diversity of available RNA binders.

Methods

Identifying NPs with Affinity for RNA

The 3 × 3 internal loop library (ILL) and 3 × 2 ILL RNAs were 5′-end labeled with 32P and purified as previously described. (35) Radiolabeled RNA was folded in 1× Assay Buffer (AB1; 8 mM Na2HPO4, pH 7.0, 185 mM NaCl, and 1 mM EDTA) by heating at 60 °C for 10 min followed by slowly cooling to room temperature. To the folded RNA library of interest were added MgCl2 (1 mM) and bovine serum albumin (BSA, 120 μg/mL) in a total volume of 2.5 mL. Note that the two ILLs were screened separately.

Prior to incubation with the labeled RNAs, microarrays were pre-equilibrated with 2.5 mL of AB1 supplemented with 1 mM MgCl₂ and 120 μg/mL BSA (or Assay Buffer 2; AB2) for 5 min at room temperature. After the slides were pre-equilibrated and excess buffer was removed, the folded RNAs were applied to the microarray surface and distributed evenly with a custom-cut piece of Parafilm. The slide was incubated with the RNA for ~15 min at room temperature. After incubation, the Parafilm was removed, and the slide was washed three times with 30 mL of AB2 for 30 s each. The glass slide was dried in a fume hood for 1 h, exposed to a phosphor screen overnight, and imaged using a Molecular Devices Typhoon phosphorimager. Of the 16,427 NP samples pinned to the microarray surface, 21 bound to the ILLs. This primary screen was completed with one replicate, while the validation of hits from this primary screen, carried out in the same way, was completed with n = 2.

Selection of Specific RNA Binders via Competition with Oligonucleotides: Two-Dimensional Combinatorial Screening (2DCS)

Microarrays were prepared as described previously.(6) In brief, molten agarose (2 mL, 1% (w/v)) was applied to a 2.0 cm × 8.4 cm glass slide, and the agarose was allowed to dry to a thin layer at room temperature for 1.5 h. Next, 300 nL aliquots of NOC-I at the indicated concentrations were pinned onto the glass slide using Beckman Biomek NXP, and the spots were allowed to completely dry overnight. All competitor oligonucleotides (C1–C8) (Figure S3), each in an amount of 0.5 equivalents to the total moles of compounds delivered to the array and ~1000-fold more than the radiolabeled ILLs, were folded separately in 1× AB1 by heating at 60 °C for 20 min followed by cooling to room temperature on the bench top. The folded oligos were mixed together with the 5′-³²P labeled RNA ILL of interest, followed by the addition of MgCl2 and BSA to final concentrations of 1 mM and 120 μg/mL, respectively, in a total volume of 600 μL. The glass slide was pre-equilibrated with AB2 for 5 min after which the excess buffer was removed. The folded oligonucleotide mixture was then applied to the surface and incubated with the array for 15 min at room temperature. The glass slide was then washed three times with AB2, dried in the fume hood for 1 h, exposed to phosphor screen overnight, and t imaged using Molecular Devices Typhoon phosphorimager (n = 1).

Preparation of Selected RNAs for Sequencing Analysis

The sequencing library was prepared using a previously described method.(22) Briefly, bound RNAs from the 2DCS selection were excised, treated with RQ DNase I (Promega), and reverse transcribed using AMV Reverse Transcriptase (New England BiioLabs (NEB)) per manufacturer’s recommended protocol. An aliquot of the RT reaction was carried forward to PCR amplification using a Taq DNA polymerase. Here, to 20 μL of the RT reaction was added 6 μL of 10× PCR buffer (500 mM KCl and 100 mM Tris–HCl, pH 8.3), 4 μL of 100 μM forward primer including a barcode (5′CCATCTCATCCCTGCGTGTCTCCGACTCAGXXXXXXXXXXGATGGGAGAGGGTTTAAT, where X represents unique barcode and GAT is the barcode adapter), 2 μL of 100 μM reverse primer (5′-CCTCTCTATGGGCAGTCGGTGATCCTTGCGGATCCAAT), 0.6 μL of 250 mM MgCl₂, and 2 μL of Taq DNA polymerase (3 mg/mL). A two-step cycle was performed to amplify the DNA: 95 °C for 1 min and 72 °C for 1 min Aliquots of PCR products were removed every two cycle to confirm that a background spot where the compound was not delivered did not amplify. The PCR product thus obtained (cycle 14) was purified using denaturing 12% polyacrylamide gel electrophoresis (dPAGE). The purity of purified cDNA was confirmed via a 2100 Bioanalyzer (Agilent Technologies). Samples were mixed in equal amounts and sequenced using an Ion Proton deep sequencer using PI chips (Thermo Fisher Scientific).

Statistical Analysis of RNA-seq Data to Identify the RNA Motifs Preferred by NOC-I: High-Throughput Structure–Activity Relationships Through Sequencing (HiT-StARTS)

HiT-StARTS analysis was completed using a previously described method. (22) Briefly, a shell script was used to extract randomized RNA sequence from fastq sequencing files generated by Scripps’s Genomic Core. The frequency of each member of the RNA library was tabulated and compared to the starting library. To determine if the difference in the frequency of an RNA in the selected mixture and its frequency in the starting library was statistically significant, a pooled population comparison (Z_obs) was calculated using eqs 1 and 2. (22)

$$\varphi =\frac{n_1{p}_1+n_2{p}_2}{n_1+n_2}$$ (eq.1)

$$Z_{obs}=\frac{\left(p1-p2\right)}{\sqrt{\varphi \left(1-\varphi \right)\left(\frac{1}{n_1}+\frac{1}{n_2}\right)}}$$ (eq.2)

where n1 is the size of population 1 (number of reads for a selected RNA); n2 is the size of population 2 (number of reads for the same RNA from sequencing of the starting library); p1 is the observed proportion of population 1 (number of reads for a selected RNA divided by the total number of reads); and p2 is the observed proportion for population 2 (number of reads for the same RNA divided by the total number of reads in the starting library).

Inhibition of In Vitro Dicer Processing of Wild-Type and Mutant pre-miR-18a

The wild-type and mutant (A bulge converted to an AU pair) pre-miR-18 RNAs (Figure S6) were in vitro transcribed from the corresponding DNA templates as previously described. (13) After purification by polyacrylamide gel electrophoresis (PAGE), they were 5′-end labeled using [32P]-ATP and T4 polynucleotide kinase and purified by PAGE as previously described. (35) The radiolabeled RNA of interest was folded in 1× AB1 by heating at 60 °C for 5 min followed by cooling to room temperature. The samples were then supplemented with 5 mM MgCl₂ and NOC-I at the indicated concentrations. After incubating at room temperature for 30 min, Dicer enzyme was added to a final concentration of 5 ng/μL, and the reactions were incubated for an additional 1 h at 37 °C. The fragments were separated on a denaturing 15% polyacrylamide gel and visualized by phosphorimaging. Each cleavage band was quantified using BioRad’s QuantityOne software and normalized to the full-length RNA.

Measuring the Affinity of NOC-I for RNA 3D Folds: A Direct, In-Solution Fluorescence Binding Assay

Dissociation constants were measured by using an in-solution, fluorescence-based assay as described previously.(35) Briefly, a selected RNA or base pair control was folded in 1× AB1 containing 40 μg/mL BSA at 90 °C for 30 s and then allowed to cool slowly to room temperature. NOC-I was then added to a final concentration of 50 nM. Serial dilutions (1:1), starting from a concentration of 100 μM of RNA, were then completed in 1× AB1 containing 40 μg/mL BSA and 50 nM NOC-I. The solutions were incubated for 20 min at room temperature and then transferred to a well of a black 384-well plate. Fluorescence intensity was measured on a Tecan-Safire II plate reader. Each measurement was completed in triplicate (independent experiments) and the results averaged. All data were fit according to eq 3: (35)

$$I=I_0+0.5\Delta ϵ\left(\left({\left[FL\right]}_0+{\left[RNA\right]}_0+K_t\right)-{\left({\left({\left[FL\right]}_0+{\left[RNA\right]}_0+K_t\right)}^2-4{\left[FL\right]}_0{\left[RNA\right]}_0\right)}^{0.5}\right)$$ (eq.3)

where I and I₀ are the observed fluorescence and initial fluorescence intensity in the presence and absence of RNA, Δε is the difference between the fluorescence intensity in the absence and presence of infinite RNA concentration, [FL]₀ and [RNA]₀ are the concentrations of the small molecule and RNA, respectively, and K_t is the dissociation constant.

Cell Culture

DU-145 cells were cultured in growth medium comprising RPMI 1640 medium (Corning) supplemented with 10% (v/v) fetal bovine serum (FBS; Sigma-Aldrich) and 1% (v/v) antimycotic-antibiotic solution (Corning) at 37 °C and 5% CO2.

RT-qPCR Analysis of Pri-, Pre-, and Mature miR-18a

For RT-qPCR analysis of NOC-I’s effect on miR-18a biogenesis, DU-145 cells were seeded in growth medium into 12-well plates (100 000 cells/well) and were allowed to adhere for 12 h. The cells were then treated with NOC-I at 2, 0.2, and 0.02 μM or DMSO (0.2% final concentration in all samples; vehicle), each prepared in a growth medium, for 48 h. After the treatment period, total RNA was extracted using a Zymo Quick-RNA Miniprep Kit per the manufacturer’s protocol, including DNase treatment. Total RNA concentration was quantified by a Thermo NanoDrop 2000C spectrophotometer.

Reverse transcription of mature miR-18a (200 ng of total RNA) was completed using a miScript II Kit (Qiagen) per the manufacturer’s protocol using the provided High Spec Buffer. To measure pre-miR-18a and primary transcript levels of the 17/92 cluster, reverse transcription was completed with a qScript cDNA Synthesis Kit (Quanta Bio) on 500 ng of total RNA per the manufacturer’s protocol. Quantitative PCR (qPCR) was completed by adding a 20 ng of cDNA generated for mature miR-18a and 100 ng of cDNA generated from pre-miR-18a or pri-miR-17/92 to 2× Power SYBR Master Mix (Applied Biosciences) and 500 nM of each primer (Table S1) in a final volume of 36 μL. Next, 10 μL aliquots of each sample were plated as three technical replicates into MicroAmp Optical 384-well reaction plates (Applied Biosciences) and analyzed using a QuantStudio 5 RT-qPCR thermocycler (Applied Biosciences) and the Comparative Ct with Melt program, modified for the 10 μL/well volume. Relative expression was calculated by the ΔΔCt method as described previously. (36)

Mapping NOC-I’s Binding Site in Cells with ASO-Bind-Map

DU-145 cells (~60% confluency) were treated with vehicle (DMSO) or NOC-I at ~60% overnight followed by transfection with a gapmer ASO (200 nM) of interest with Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer’s instructions. Mock samples were generated by treating cells with transfection reagent lacking oligonucleotide. RNA extraction, cDNA synthesis, and RT-qPCR were performed as described above after 48 h of treatment. Primers for pre-miR-18a are provided in Table S1.

Assessing Apoptosis by a Caspase 3/7 Assay

Measurement of apoptosis was carried out as previously described. (22) Briefly, white 96-well clear bottom tissue culture plates (Corning; catalog no. CLS3903) were seeded with DU-145 cells (10 000 cells/well), which were allowed to adhere for 12 h. Once adhered, the cells were treated with NOC-I prepared in growth medium to final concentrations of 2, 0.2, and 0.02 μM for 48 h. After 48 h, Caspase 3/7 Glo (Promega) reagent was added to the cells per the manufacturer’s protocol, and luminescence was measured using a SpectraMax M5 plate reader (Molecular Devices) with an integration time of 500 ms.

miRNA Profiling

The selectivity of NOC-I miRnome-wide was assessed by profiling its effect on 373 expressed miRNAs expressed in DU-145 cells by RT-qPCR. Briefly, DU-145 cells were plated into 6-well plates (400 000 cells/well) and allowed to grow until they reached ~80% confluency. The cells were treated with 2 μM NOC-I prepared in a growth medium for 12 h. [Note: the cells were not treated for 48 h due to the induction of apoptosis which causes global degradation of all cellular RNAs.] Total RNA was then harvested using a Qiagen miRNeasy RNA Extraction Kit with DNase I treatment per the manufacturer’s protocol. Reverse transcription was completed using 2 μg of total RNA and the miScript II RT Kit (Qiagen) per the manufacturer’s protocol in a total volume of 20 μL. The RT reaction (20 μL) was aliquoted amongst separate reactions for each of the 373 miRNA scontaining Power SYBR Mast Mix and primers (500 nM each). Both qPCR amplification and data analysis were completed as described previously.(24)